hibernate.c 35 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031
  1. /* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
  2. * Copyright (c) 2007-2013, The Tor Project, Inc. */
  3. /* See LICENSE for licensing information */
  4. /**
  5. * \file hibernate.c
  6. * \brief Functions to close listeners, stop allowing new circuits,
  7. * etc in preparation for closing down or going dormant; and to track
  8. * bandwidth and time intervals to know when to hibernate and when to
  9. * stop hibernating.
  10. **/
  11. /*
  12. hibernating, phase 1:
  13. - send destroy in response to create cells
  14. - send end (policy failed) in response to begin cells
  15. - close an OR conn when it has no circuits
  16. hibernating, phase 2:
  17. (entered when bandwidth hard limit reached)
  18. - close all OR/AP/exit conns)
  19. */
  20. #define HIBERNATE_PRIVATE
  21. #include "or.h"
  22. #include "channel.h"
  23. #include "channeltls.h"
  24. #include "config.h"
  25. #include "connection.h"
  26. #include "connection_edge.h"
  27. #include "hibernate.h"
  28. #include "main.h"
  29. #include "router.h"
  30. #include "statefile.h"
  31. extern long stats_n_seconds_working; /* published uptime */
  32. /** Are we currently awake, asleep, running out of bandwidth, or shutting
  33. * down? */
  34. static hibernate_state_t hibernate_state = HIBERNATE_STATE_INITIAL;
  35. /** If are hibernating, when do we plan to wake up? Set to 0 if we
  36. * aren't hibernating. */
  37. static time_t hibernate_end_time = 0;
  38. /** If we are shutting down, when do we plan finally exit? Set to 0 if
  39. * we aren't shutting down. */
  40. static time_t shutdown_time = 0;
  41. /** Possible accounting periods. */
  42. typedef enum {
  43. UNIT_MONTH=1, UNIT_WEEK=2, UNIT_DAY=3,
  44. } time_unit_t;
  45. /* Fields for accounting logic. Accounting overview:
  46. *
  47. * Accounting is designed to ensure that no more than N bytes are sent in
  48. * either direction over a given interval (currently, one month, one week, or
  49. * one day) We could
  50. * try to do this by choking our bandwidth to a trickle, but that
  51. * would make our streams useless. Instead, we estimate what our
  52. * bandwidth usage will be, and guess how long we'll be able to
  53. * provide that much bandwidth before hitting our limit. We then
  54. * choose a random time within the accounting interval to come up (so
  55. * that we don't get 50 Tors running on the 1st of the month and none
  56. * on the 30th).
  57. *
  58. * Each interval runs as follows:
  59. *
  60. * 1. We guess our bandwidth usage, based on how much we used
  61. * last time. We choose a "wakeup time" within the interval to come up.
  62. * 2. Until the chosen wakeup time, we hibernate.
  63. * 3. We come up at the wakeup time, and provide bandwidth until we are
  64. * "very close" to running out.
  65. * 4. Then we go into low-bandwidth mode, and stop accepting new
  66. * connections, but provide bandwidth until we run out.
  67. * 5. Then we hibernate until the end of the interval.
  68. *
  69. * If the interval ends before we run out of bandwidth, we go back to
  70. * step one.
  71. */
  72. /** How many bytes have we read in this accounting interval? */
  73. static uint64_t n_bytes_read_in_interval = 0;
  74. /** How many bytes have we written in this accounting interval? */
  75. static uint64_t n_bytes_written_in_interval = 0;
  76. /** How many seconds have we been running this interval? */
  77. static uint32_t n_seconds_active_in_interval = 0;
  78. /** How many seconds were we active in this interval before we hit our soft
  79. * limit? */
  80. static int n_seconds_to_hit_soft_limit = 0;
  81. /** When in this interval was the soft limit hit. */
  82. static time_t soft_limit_hit_at = 0;
  83. /** How many bytes had we read/written when we hit the soft limit? */
  84. static uint64_t n_bytes_at_soft_limit = 0;
  85. /** When did this accounting interval start? */
  86. static time_t interval_start_time = 0;
  87. /** When will this accounting interval end? */
  88. static time_t interval_end_time = 0;
  89. /** How far into the accounting interval should we hibernate? */
  90. static time_t interval_wakeup_time = 0;
  91. /** How much bandwidth do we 'expect' to use per minute? (0 if we have no
  92. * info from the last period.) */
  93. static uint64_t expected_bandwidth_usage = 0;
  94. /** What unit are we using for our accounting? */
  95. static time_unit_t cfg_unit = UNIT_MONTH;
  96. /** How many days,hours,minutes into each unit does our accounting interval
  97. * start? */
  98. /** @{ */
  99. static int cfg_start_day = 0,
  100. cfg_start_hour = 0,
  101. cfg_start_min = 0;
  102. /** @} */
  103. static void reset_accounting(time_t now);
  104. static int read_bandwidth_usage(void);
  105. static time_t start_of_accounting_period_after(time_t now);
  106. static time_t start_of_accounting_period_containing(time_t now);
  107. static void accounting_set_wakeup_time(void);
  108. /* ************
  109. * Functions for bandwidth accounting.
  110. * ************/
  111. /** Configure accounting start/end time settings based on
  112. * options->AccountingStart. Return 0 on success, -1 on failure. If
  113. * <b>validate_only</b> is true, do not change the current settings. */
  114. int
  115. accounting_parse_options(const or_options_t *options, int validate_only)
  116. {
  117. time_unit_t unit;
  118. int ok, idx;
  119. long d,h,m;
  120. smartlist_t *items;
  121. const char *v = options->AccountingStart;
  122. const char *s;
  123. char *cp;
  124. if (!v) {
  125. if (!validate_only) {
  126. cfg_unit = UNIT_MONTH;
  127. cfg_start_day = 1;
  128. cfg_start_hour = 0;
  129. cfg_start_min = 0;
  130. }
  131. return 0;
  132. }
  133. items = smartlist_new();
  134. smartlist_split_string(items, v, NULL,
  135. SPLIT_SKIP_SPACE|SPLIT_IGNORE_BLANK,0);
  136. if (smartlist_len(items)<2) {
  137. log_warn(LD_CONFIG, "Too few arguments to AccountingStart");
  138. goto err;
  139. }
  140. s = smartlist_get(items,0);
  141. if (0==strcasecmp(s, "month")) {
  142. unit = UNIT_MONTH;
  143. } else if (0==strcasecmp(s, "week")) {
  144. unit = UNIT_WEEK;
  145. } else if (0==strcasecmp(s, "day")) {
  146. unit = UNIT_DAY;
  147. } else {
  148. log_warn(LD_CONFIG,
  149. "Unrecognized accounting unit '%s': only 'month', 'week',"
  150. " and 'day' are supported.", s);
  151. goto err;
  152. }
  153. switch (unit) {
  154. case UNIT_WEEK:
  155. d = tor_parse_long(smartlist_get(items,1), 10, 1, 7, &ok, NULL);
  156. if (!ok) {
  157. log_warn(LD_CONFIG, "Weekly accounting must begin on a day between "
  158. "1 (Monday) and 7 (Sunday)");
  159. goto err;
  160. }
  161. break;
  162. case UNIT_MONTH:
  163. d = tor_parse_long(smartlist_get(items,1), 10, 1, 28, &ok, NULL);
  164. if (!ok) {
  165. log_warn(LD_CONFIG, "Monthly accounting must begin on a day between "
  166. "1 and 28");
  167. goto err;
  168. }
  169. break;
  170. case UNIT_DAY:
  171. d = 0;
  172. break;
  173. /* Coverity dislikes unreachable default cases; some compilers warn on
  174. * switch statements missing a case. Tell Coverity not to worry. */
  175. /* coverity[dead_error_begin] */
  176. default:
  177. tor_assert(0);
  178. }
  179. idx = unit==UNIT_DAY?1:2;
  180. if (smartlist_len(items) != (idx+1)) {
  181. log_warn(LD_CONFIG,"Accounting unit '%s' requires %d argument%s.",
  182. s, idx, (idx>1)?"s":"");
  183. goto err;
  184. }
  185. s = smartlist_get(items, idx);
  186. h = tor_parse_long(s, 10, 0, 23, &ok, &cp);
  187. if (!ok) {
  188. log_warn(LD_CONFIG,"Accounting start time not parseable: bad hour.");
  189. goto err;
  190. }
  191. if (!cp || *cp!=':') {
  192. log_warn(LD_CONFIG,
  193. "Accounting start time not parseable: not in HH:MM format");
  194. goto err;
  195. }
  196. m = tor_parse_long(cp+1, 10, 0, 59, &ok, &cp);
  197. if (!ok) {
  198. log_warn(LD_CONFIG, "Accounting start time not parseable: bad minute");
  199. goto err;
  200. }
  201. if (!cp || *cp!='\0') {
  202. log_warn(LD_CONFIG,
  203. "Accounting start time not parseable: not in HH:MM format");
  204. goto err;
  205. }
  206. if (!validate_only) {
  207. cfg_unit = unit;
  208. cfg_start_day = (int)d;
  209. cfg_start_hour = (int)h;
  210. cfg_start_min = (int)m;
  211. }
  212. SMARTLIST_FOREACH(items, char *, item, tor_free(item));
  213. smartlist_free(items);
  214. return 0;
  215. err:
  216. SMARTLIST_FOREACH(items, char *, item, tor_free(item));
  217. smartlist_free(items);
  218. return -1;
  219. }
  220. /** If we want to manage the accounting system and potentially
  221. * hibernate, return 1, else return 0.
  222. */
  223. int
  224. accounting_is_enabled(const or_options_t *options)
  225. {
  226. if (options->AccountingMax)
  227. return 1;
  228. return 0;
  229. }
  230. /** If accounting is enabled, return how long (in seconds) this
  231. * interval lasts. */
  232. int
  233. accounting_get_interval_length(void)
  234. {
  235. return (int)(interval_end_time - interval_start_time);
  236. }
  237. /** Return the time at which the current accounting interval will end. */
  238. time_t
  239. accounting_get_end_time(void)
  240. {
  241. return interval_end_time;
  242. }
  243. /** Called from main.c to tell us that <b>seconds</b> seconds have
  244. * passed, <b>n_read</b> bytes have been read, and <b>n_written</b>
  245. * bytes have been written. */
  246. void
  247. accounting_add_bytes(size_t n_read, size_t n_written, int seconds)
  248. {
  249. n_bytes_read_in_interval += n_read;
  250. n_bytes_written_in_interval += n_written;
  251. /* If we haven't been called in 10 seconds, we're probably jumping
  252. * around in time. */
  253. n_seconds_active_in_interval += (seconds < 10) ? seconds : 0;
  254. }
  255. /** If get_end, return the end of the accounting period that contains
  256. * the time <b>now</b>. Else, return the start of the accounting
  257. * period that contains the time <b>now</b> */
  258. static time_t
  259. edge_of_accounting_period_containing(time_t now, int get_end)
  260. {
  261. int before;
  262. struct tm tm;
  263. tor_localtime_r(&now, &tm);
  264. /* Set 'before' to true iff the current time is before the hh:mm
  265. * changeover time for today. */
  266. before = tm.tm_hour < cfg_start_hour ||
  267. (tm.tm_hour == cfg_start_hour && tm.tm_min < cfg_start_min);
  268. /* Dispatch by unit. First, find the start day of the given period;
  269. * then, if get_end is true, increment to the end day. */
  270. switch (cfg_unit)
  271. {
  272. case UNIT_MONTH: {
  273. /* If this is before the Nth, we want the Nth of last month. */
  274. if (tm.tm_mday < cfg_start_day ||
  275. (tm.tm_mday < cfg_start_day && before)) {
  276. --tm.tm_mon;
  277. }
  278. /* Otherwise, the month is correct. */
  279. tm.tm_mday = cfg_start_day;
  280. if (get_end)
  281. ++tm.tm_mon;
  282. break;
  283. }
  284. case UNIT_WEEK: {
  285. /* What is the 'target' day of the week in struct tm format? (We
  286. say Sunday==7; struct tm says Sunday==0.) */
  287. int wday = cfg_start_day % 7;
  288. /* How many days do we subtract from today to get to the right day? */
  289. int delta = (7+tm.tm_wday-wday)%7;
  290. /* If we are on the right day, but the changeover hasn't happened yet,
  291. * then subtract a whole week. */
  292. if (delta == 0 && before)
  293. delta = 7;
  294. tm.tm_mday -= delta;
  295. if (get_end)
  296. tm.tm_mday += 7;
  297. break;
  298. }
  299. case UNIT_DAY:
  300. if (before)
  301. --tm.tm_mday;
  302. if (get_end)
  303. ++tm.tm_mday;
  304. break;
  305. default:
  306. tor_assert(0);
  307. }
  308. tm.tm_hour = cfg_start_hour;
  309. tm.tm_min = cfg_start_min;
  310. tm.tm_sec = 0;
  311. tm.tm_isdst = -1; /* Autodetect DST */
  312. return mktime(&tm);
  313. }
  314. /** Return the start of the accounting period containing the time
  315. * <b>now</b>. */
  316. static time_t
  317. start_of_accounting_period_containing(time_t now)
  318. {
  319. return edge_of_accounting_period_containing(now, 0);
  320. }
  321. /** Return the start of the accounting period that comes after the one
  322. * containing the time <b>now</b>. */
  323. static time_t
  324. start_of_accounting_period_after(time_t now)
  325. {
  326. return edge_of_accounting_period_containing(now, 1);
  327. }
  328. /** Return the length of the accounting period containing the time
  329. * <b>now</b>. */
  330. static long
  331. length_of_accounting_period_containing(time_t now)
  332. {
  333. return edge_of_accounting_period_containing(now, 1) -
  334. edge_of_accounting_period_containing(now, 0);
  335. }
  336. /** Initialize the accounting subsystem. */
  337. void
  338. configure_accounting(time_t now)
  339. {
  340. time_t s_now;
  341. /* Try to remember our recorded usage. */
  342. if (!interval_start_time)
  343. read_bandwidth_usage(); /* If we fail, we'll leave values at zero, and
  344. * reset below.*/
  345. s_now = start_of_accounting_period_containing(now);
  346. if (!interval_start_time) {
  347. /* We didn't have recorded usage; Start a new interval. */
  348. log_info(LD_ACCT, "Starting new accounting interval.");
  349. reset_accounting(now);
  350. } else if (s_now == interval_start_time) {
  351. log_info(LD_ACCT, "Continuing accounting interval.");
  352. /* We are in the interval we thought we were in. Do nothing.*/
  353. interval_end_time = start_of_accounting_period_after(interval_start_time);
  354. } else {
  355. long duration =
  356. length_of_accounting_period_containing(interval_start_time);
  357. double delta = ((double)(s_now - interval_start_time)) / duration;
  358. if (-0.50 <= delta && delta <= 0.50) {
  359. /* The start of the period is now a little later or earlier than we
  360. * remembered. That's fine; we might lose some bytes we could otherwise
  361. * have written, but better to err on the side of obeying people's
  362. * accounting settings. */
  363. log_info(LD_ACCT, "Accounting interval moved by %.02f%%; "
  364. "that's fine.", delta*100);
  365. interval_end_time = start_of_accounting_period_after(now);
  366. } else if (delta >= 0.99) {
  367. /* This is the regular time-moved-forward case; don't be too noisy
  368. * about it or people will complain */
  369. log_info(LD_ACCT, "Accounting interval elapsed; starting a new one");
  370. reset_accounting(now);
  371. } else {
  372. log_warn(LD_ACCT,
  373. "Mismatched accounting interval: moved by %.02f%%. "
  374. "Starting a fresh one.", delta*100);
  375. reset_accounting(now);
  376. }
  377. }
  378. accounting_set_wakeup_time();
  379. }
  380. /** Set expected_bandwidth_usage based on how much we sent/received
  381. * per minute last interval (if we were up for at least 30 minutes),
  382. * or based on our declared bandwidth otherwise. */
  383. static void
  384. update_expected_bandwidth(void)
  385. {
  386. uint64_t expected;
  387. const or_options_t *options= get_options();
  388. uint64_t max_configured = (options->RelayBandwidthRate > 0 ?
  389. options->RelayBandwidthRate :
  390. options->BandwidthRate) * 60;
  391. #define MIN_TIME_FOR_MEASUREMENT (1800)
  392. if (soft_limit_hit_at > interval_start_time && n_bytes_at_soft_limit &&
  393. (soft_limit_hit_at - interval_start_time) > MIN_TIME_FOR_MEASUREMENT) {
  394. /* If we hit our soft limit last time, only count the bytes up to that
  395. * time. This is a better predictor of our actual bandwidth than
  396. * considering the entirety of the last interval, since we likely started
  397. * using bytes very slowly once we hit our soft limit. */
  398. expected = n_bytes_at_soft_limit /
  399. (soft_limit_hit_at - interval_start_time);
  400. expected /= 60;
  401. } else if (n_seconds_active_in_interval >= MIN_TIME_FOR_MEASUREMENT) {
  402. /* Otherwise, we either measured enough time in the last interval but
  403. * never hit our soft limit, or we're using a state file from a Tor that
  404. * doesn't know to store soft-limit info. Just take rate at which
  405. * we were reading/writing in the last interval as our expected rate.
  406. */
  407. uint64_t used = MAX(n_bytes_written_in_interval,
  408. n_bytes_read_in_interval);
  409. expected = used / (n_seconds_active_in_interval / 60);
  410. } else {
  411. /* If we haven't gotten enough data last interval, set 'expected'
  412. * to 0. This will set our wakeup to the start of the interval.
  413. * Next interval, we'll choose our starting time based on how much
  414. * we sent this interval.
  415. */
  416. expected = 0;
  417. }
  418. if (expected > max_configured)
  419. expected = max_configured;
  420. expected_bandwidth_usage = expected;
  421. }
  422. /** Called at the start of a new accounting interval: reset our
  423. * expected bandwidth usage based on what happened last time, set up
  424. * the start and end of the interval, and clear byte/time totals.
  425. */
  426. static void
  427. reset_accounting(time_t now)
  428. {
  429. log_info(LD_ACCT, "Starting new accounting interval.");
  430. update_expected_bandwidth();
  431. interval_start_time = start_of_accounting_period_containing(now);
  432. interval_end_time = start_of_accounting_period_after(interval_start_time);
  433. n_bytes_read_in_interval = 0;
  434. n_bytes_written_in_interval = 0;
  435. n_seconds_active_in_interval = 0;
  436. n_bytes_at_soft_limit = 0;
  437. soft_limit_hit_at = 0;
  438. n_seconds_to_hit_soft_limit = 0;
  439. }
  440. /** Return true iff we should save our bandwidth usage to disk. */
  441. static INLINE int
  442. time_to_record_bandwidth_usage(time_t now)
  443. {
  444. /* Note every 600 sec */
  445. #define NOTE_INTERVAL (600)
  446. /* Or every 20 megabytes */
  447. #define NOTE_BYTES 20*(1024*1024)
  448. static uint64_t last_read_bytes_noted = 0;
  449. static uint64_t last_written_bytes_noted = 0;
  450. static time_t last_time_noted = 0;
  451. if (last_time_noted + NOTE_INTERVAL <= now ||
  452. last_read_bytes_noted + NOTE_BYTES <= n_bytes_read_in_interval ||
  453. last_written_bytes_noted + NOTE_BYTES <= n_bytes_written_in_interval ||
  454. (interval_end_time && interval_end_time <= now)) {
  455. last_time_noted = now;
  456. last_read_bytes_noted = n_bytes_read_in_interval;
  457. last_written_bytes_noted = n_bytes_written_in_interval;
  458. return 1;
  459. }
  460. return 0;
  461. }
  462. /** Invoked once per second. Checks whether it is time to hibernate,
  463. * record bandwidth used, etc. */
  464. void
  465. accounting_run_housekeeping(time_t now)
  466. {
  467. if (now >= interval_end_time) {
  468. configure_accounting(now);
  469. }
  470. if (time_to_record_bandwidth_usage(now)) {
  471. if (accounting_record_bandwidth_usage(now, get_or_state())) {
  472. log_warn(LD_FS, "Couldn't record bandwidth usage to disk.");
  473. }
  474. }
  475. }
  476. /** Based on our interval and our estimated bandwidth, choose a
  477. * deterministic (but random-ish) time to wake up. */
  478. static void
  479. accounting_set_wakeup_time(void)
  480. {
  481. char digest[DIGEST_LEN];
  482. crypto_digest_t *d_env;
  483. uint64_t time_to_exhaust_bw;
  484. int time_to_consider;
  485. if (! server_identity_key_is_set()) {
  486. if (init_keys() < 0) {
  487. log_err(LD_BUG, "Error initializing keys");
  488. tor_assert(0);
  489. }
  490. }
  491. if (server_identity_key_is_set()) {
  492. char buf[ISO_TIME_LEN+1];
  493. format_iso_time(buf, interval_start_time);
  494. crypto_pk_get_digest(get_server_identity_key(), digest);
  495. d_env = crypto_digest_new();
  496. crypto_digest_add_bytes(d_env, buf, ISO_TIME_LEN);
  497. crypto_digest_add_bytes(d_env, digest, DIGEST_LEN);
  498. crypto_digest_get_digest(d_env, digest, DIGEST_LEN);
  499. crypto_digest_free(d_env);
  500. } else {
  501. crypto_rand(digest, DIGEST_LEN);
  502. }
  503. if (!expected_bandwidth_usage) {
  504. char buf1[ISO_TIME_LEN+1];
  505. char buf2[ISO_TIME_LEN+1];
  506. format_local_iso_time(buf1, interval_start_time);
  507. format_local_iso_time(buf2, interval_end_time);
  508. interval_wakeup_time = interval_start_time;
  509. log_notice(LD_ACCT,
  510. "Configured hibernation. This interval begins at %s "
  511. "and ends at %s. We have no prior estimate for bandwidth, so "
  512. "we will start out awake and hibernate when we exhaust our quota.",
  513. buf1, buf2);
  514. return;
  515. }
  516. time_to_exhaust_bw =
  517. (get_options()->AccountingMax/expected_bandwidth_usage)*60;
  518. if (time_to_exhaust_bw > INT_MAX) {
  519. time_to_exhaust_bw = INT_MAX;
  520. time_to_consider = 0;
  521. } else {
  522. time_to_consider = accounting_get_interval_length() -
  523. (int)time_to_exhaust_bw;
  524. }
  525. if (time_to_consider<=0) {
  526. interval_wakeup_time = interval_start_time;
  527. } else {
  528. /* XXX can we simplify this just by picking a random (non-deterministic)
  529. * time to be up? If we go down and come up, then we pick a new one. Is
  530. * that good enough? -RD */
  531. /* This is not a perfectly unbiased conversion, but it is good enough:
  532. * in the worst case, the first half of the day is 0.06 percent likelier
  533. * to be chosen than the last half. */
  534. interval_wakeup_time = interval_start_time +
  535. (get_uint32(digest) % time_to_consider);
  536. }
  537. {
  538. char buf1[ISO_TIME_LEN+1];
  539. char buf2[ISO_TIME_LEN+1];
  540. char buf3[ISO_TIME_LEN+1];
  541. char buf4[ISO_TIME_LEN+1];
  542. time_t down_time;
  543. if (interval_wakeup_time+time_to_exhaust_bw > TIME_MAX)
  544. down_time = TIME_MAX;
  545. else
  546. down_time = (time_t)(interval_wakeup_time+time_to_exhaust_bw);
  547. if (down_time>interval_end_time)
  548. down_time = interval_end_time;
  549. format_local_iso_time(buf1, interval_start_time);
  550. format_local_iso_time(buf2, interval_wakeup_time);
  551. format_local_iso_time(buf3, down_time);
  552. format_local_iso_time(buf4, interval_end_time);
  553. log_notice(LD_ACCT,
  554. "Configured hibernation. This interval began at %s; "
  555. "the scheduled wake-up time %s %s; "
  556. "we expect%s to exhaust our quota for this interval around %s; "
  557. "the next interval begins at %s (all times local)",
  558. buf1,
  559. time(NULL)<interval_wakeup_time?"is":"was", buf2,
  560. time(NULL)<down_time?"":"ed", buf3,
  561. buf4);
  562. }
  563. }
  564. /* This rounds 0 up to 1000, but that's actually a feature. */
  565. #define ROUND_UP(x) (((x) + 0x3ff) & ~0x3ff)
  566. /** Save all our bandwidth tracking information to disk. Return 0 on
  567. * success, -1 on failure. */
  568. int
  569. accounting_record_bandwidth_usage(time_t now, or_state_t *state)
  570. {
  571. /* Just update the state */
  572. state->AccountingIntervalStart = interval_start_time;
  573. state->AccountingBytesReadInInterval = ROUND_UP(n_bytes_read_in_interval);
  574. state->AccountingBytesWrittenInInterval =
  575. ROUND_UP(n_bytes_written_in_interval);
  576. state->AccountingSecondsActive = n_seconds_active_in_interval;
  577. state->AccountingExpectedUsage = expected_bandwidth_usage;
  578. state->AccountingSecondsToReachSoftLimit = n_seconds_to_hit_soft_limit;
  579. state->AccountingSoftLimitHitAt = soft_limit_hit_at;
  580. state->AccountingBytesAtSoftLimit = n_bytes_at_soft_limit;
  581. or_state_mark_dirty(state,
  582. now+(get_options()->AvoidDiskWrites ? 7200 : 60));
  583. return 0;
  584. }
  585. #undef ROUND_UP
  586. /** Read stored accounting information from disk. Return 0 on success;
  587. * return -1 and change nothing on failure. */
  588. static int
  589. read_bandwidth_usage(void)
  590. {
  591. or_state_t *state = get_or_state();
  592. {
  593. char *fname = get_datadir_fname("bw_accounting");
  594. unlink(fname);
  595. tor_free(fname);
  596. }
  597. if (!state)
  598. return -1;
  599. log_info(LD_ACCT, "Reading bandwidth accounting data from state file");
  600. n_bytes_read_in_interval = state->AccountingBytesReadInInterval;
  601. n_bytes_written_in_interval = state->AccountingBytesWrittenInInterval;
  602. n_seconds_active_in_interval = state->AccountingSecondsActive;
  603. interval_start_time = state->AccountingIntervalStart;
  604. expected_bandwidth_usage = state->AccountingExpectedUsage;
  605. /* Older versions of Tor (before 0.2.2.17-alpha or so) didn't generate these
  606. * fields. If you switch back and forth, you might get an
  607. * AccountingSoftLimitHitAt value from long before the most recent
  608. * interval_start_time. If that's so, then ignore the softlimit-related
  609. * values. */
  610. if (state->AccountingSoftLimitHitAt > interval_start_time) {
  611. soft_limit_hit_at = state->AccountingSoftLimitHitAt;
  612. n_bytes_at_soft_limit = state->AccountingBytesAtSoftLimit;
  613. n_seconds_to_hit_soft_limit = state->AccountingSecondsToReachSoftLimit;
  614. } else {
  615. soft_limit_hit_at = 0;
  616. n_bytes_at_soft_limit = 0;
  617. n_seconds_to_hit_soft_limit = 0;
  618. }
  619. {
  620. char tbuf1[ISO_TIME_LEN+1];
  621. char tbuf2[ISO_TIME_LEN+1];
  622. format_iso_time(tbuf1, state->LastWritten);
  623. format_iso_time(tbuf2, state->AccountingIntervalStart);
  624. log_info(LD_ACCT,
  625. "Successfully read bandwidth accounting info from state written at %s "
  626. "for interval starting at %s. We have been active for %lu seconds in "
  627. "this interval. At the start of the interval, we expected to use "
  628. "about %lu KB per second. ("U64_FORMAT" bytes read so far, "
  629. U64_FORMAT" bytes written so far)",
  630. tbuf1, tbuf2,
  631. (unsigned long)n_seconds_active_in_interval,
  632. (unsigned long)(expected_bandwidth_usage*1024/60),
  633. U64_PRINTF_ARG(n_bytes_read_in_interval),
  634. U64_PRINTF_ARG(n_bytes_written_in_interval));
  635. }
  636. return 0;
  637. }
  638. /** Return true iff we have sent/received all the bytes we are willing
  639. * to send/receive this interval. */
  640. static int
  641. hibernate_hard_limit_reached(void)
  642. {
  643. uint64_t hard_limit = get_options()->AccountingMax;
  644. if (!hard_limit)
  645. return 0;
  646. return n_bytes_read_in_interval >= hard_limit
  647. || n_bytes_written_in_interval >= hard_limit;
  648. }
  649. /** Return true iff we have sent/received almost all the bytes we are willing
  650. * to send/receive this interval. */
  651. static int
  652. hibernate_soft_limit_reached(void)
  653. {
  654. const uint64_t acct_max = get_options()->AccountingMax;
  655. #define SOFT_LIM_PCT (.95)
  656. #define SOFT_LIM_BYTES (500*1024*1024)
  657. #define SOFT_LIM_MINUTES (3*60)
  658. /* The 'soft limit' is a fair bit more complicated now than once it was.
  659. * We want to stop accepting connections when ALL of the following are true:
  660. * - We expect to use up the remaining bytes in under 3 hours
  661. * - We have used up 95% of our bytes.
  662. * - We have less than 500MB of bytes left.
  663. */
  664. uint64_t soft_limit = DBL_TO_U64(U64_TO_DBL(acct_max) * SOFT_LIM_PCT);
  665. if (acct_max > SOFT_LIM_BYTES && acct_max - SOFT_LIM_BYTES > soft_limit) {
  666. soft_limit = acct_max - SOFT_LIM_BYTES;
  667. }
  668. if (expected_bandwidth_usage) {
  669. const uint64_t expected_usage =
  670. expected_bandwidth_usage * SOFT_LIM_MINUTES;
  671. if (acct_max > expected_usage && acct_max - expected_usage > soft_limit)
  672. soft_limit = acct_max - expected_usage;
  673. }
  674. if (!soft_limit)
  675. return 0;
  676. return n_bytes_read_in_interval >= soft_limit
  677. || n_bytes_written_in_interval >= soft_limit;
  678. }
  679. /** Called when we get a SIGINT, or when bandwidth soft limit is
  680. * reached. Puts us into "loose hibernation": we don't accept new
  681. * connections, but we continue handling old ones. */
  682. static void
  683. hibernate_begin(hibernate_state_t new_state, time_t now)
  684. {
  685. const or_options_t *options = get_options();
  686. if (new_state == HIBERNATE_STATE_EXITING &&
  687. hibernate_state != HIBERNATE_STATE_LIVE) {
  688. log_notice(LD_GENERAL,"SIGINT received %s; exiting now.",
  689. hibernate_state == HIBERNATE_STATE_EXITING ?
  690. "a second time" : "while hibernating");
  691. tor_cleanup();
  692. exit(0);
  693. }
  694. if (new_state == HIBERNATE_STATE_LOWBANDWIDTH &&
  695. hibernate_state == HIBERNATE_STATE_LIVE) {
  696. soft_limit_hit_at = now;
  697. n_seconds_to_hit_soft_limit = n_seconds_active_in_interval;
  698. n_bytes_at_soft_limit = MAX(n_bytes_read_in_interval,
  699. n_bytes_written_in_interval);
  700. }
  701. /* close listeners. leave control listener(s). */
  702. connection_mark_all_noncontrol_listeners();
  703. /* XXX kill intro point circs */
  704. /* XXX upload rendezvous service descriptors with no intro points */
  705. if (new_state == HIBERNATE_STATE_EXITING) {
  706. log_notice(LD_GENERAL,"Interrupt: we have stopped accepting new "
  707. "connections, and will shut down in %d seconds. Interrupt "
  708. "again to exit now.", options->ShutdownWaitLength);
  709. shutdown_time = time(NULL) + options->ShutdownWaitLength;
  710. } else { /* soft limit reached */
  711. hibernate_end_time = interval_end_time;
  712. }
  713. hibernate_state = new_state;
  714. accounting_record_bandwidth_usage(now, get_or_state());
  715. or_state_mark_dirty(get_or_state(),
  716. get_options()->AvoidDiskWrites ? now+600 : 0);
  717. }
  718. /** Called when we've been hibernating and our timeout is reached. */
  719. static void
  720. hibernate_end(hibernate_state_t new_state)
  721. {
  722. tor_assert(hibernate_state == HIBERNATE_STATE_LOWBANDWIDTH ||
  723. hibernate_state == HIBERNATE_STATE_DORMANT ||
  724. hibernate_state == HIBERNATE_STATE_INITIAL);
  725. /* listeners will be relaunched in run_scheduled_events() in main.c */
  726. if (hibernate_state != HIBERNATE_STATE_INITIAL)
  727. log_notice(LD_ACCT,"Hibernation period ended. Resuming normal activity.");
  728. hibernate_state = new_state;
  729. hibernate_end_time = 0; /* no longer hibernating */
  730. stats_n_seconds_working = 0; /* reset published uptime */
  731. }
  732. /** A wrapper around hibernate_begin, for when we get SIGINT. */
  733. void
  734. hibernate_begin_shutdown(void)
  735. {
  736. hibernate_begin(HIBERNATE_STATE_EXITING, time(NULL));
  737. }
  738. /** Return true iff we are currently hibernating. */
  739. int
  740. we_are_hibernating(void)
  741. {
  742. return hibernate_state != HIBERNATE_STATE_LIVE;
  743. }
  744. /** If we aren't currently dormant, close all connections and become
  745. * dormant. */
  746. static void
  747. hibernate_go_dormant(time_t now)
  748. {
  749. connection_t *conn;
  750. if (hibernate_state == HIBERNATE_STATE_DORMANT)
  751. return;
  752. else if (hibernate_state == HIBERNATE_STATE_LOWBANDWIDTH)
  753. hibernate_state = HIBERNATE_STATE_DORMANT;
  754. else
  755. hibernate_begin(HIBERNATE_STATE_DORMANT, now);
  756. log_notice(LD_ACCT,"Going dormant. Blowing away remaining connections.");
  757. /* Close all OR/AP/exit conns. Leave dir conns because we still want
  758. * to be able to upload server descriptors so people know we're still
  759. * running, and download directories so we can detect if we're obsolete.
  760. * Leave control conns because we still want to be controllable.
  761. */
  762. while ((conn = connection_get_by_type(CONN_TYPE_OR)) ||
  763. (conn = connection_get_by_type(CONN_TYPE_AP)) ||
  764. (conn = connection_get_by_type(CONN_TYPE_EXIT))) {
  765. if (CONN_IS_EDGE(conn))
  766. connection_edge_end(TO_EDGE_CONN(conn), END_STREAM_REASON_HIBERNATING);
  767. log_info(LD_NET,"Closing conn type %d", conn->type);
  768. if (conn->type == CONN_TYPE_AP) /* send socks failure if needed */
  769. connection_mark_unattached_ap(TO_ENTRY_CONN(conn),
  770. END_STREAM_REASON_HIBERNATING);
  771. else if (conn->type == CONN_TYPE_OR) {
  772. if (TO_OR_CONN(conn)->chan) {
  773. channel_mark_for_close(TLS_CHAN_TO_BASE(TO_OR_CONN(conn)->chan));
  774. } else {
  775. connection_mark_for_close(conn);
  776. }
  777. } else
  778. connection_mark_for_close(conn);
  779. }
  780. if (now < interval_wakeup_time)
  781. hibernate_end_time = interval_wakeup_time;
  782. else
  783. hibernate_end_time = interval_end_time;
  784. accounting_record_bandwidth_usage(now, get_or_state());
  785. or_state_mark_dirty(get_or_state(),
  786. get_options()->AvoidDiskWrites ? now+600 : 0);
  787. }
  788. /** Called when hibernate_end_time has arrived. */
  789. static void
  790. hibernate_end_time_elapsed(time_t now)
  791. {
  792. char buf[ISO_TIME_LEN+1];
  793. /* The interval has ended, or it is wakeup time. Find out which. */
  794. accounting_run_housekeeping(now);
  795. if (interval_wakeup_time <= now) {
  796. /* The interval hasn't changed, but interval_wakeup_time has passed.
  797. * It's time to wake up and start being a server. */
  798. hibernate_end(HIBERNATE_STATE_LIVE);
  799. return;
  800. } else {
  801. /* The interval has changed, and it isn't time to wake up yet. */
  802. hibernate_end_time = interval_wakeup_time;
  803. format_iso_time(buf,interval_wakeup_time);
  804. if (hibernate_state != HIBERNATE_STATE_DORMANT) {
  805. /* We weren't sleeping before; we should sleep now. */
  806. log_notice(LD_ACCT,
  807. "Accounting period ended. Commencing hibernation until "
  808. "%s UTC", buf);
  809. hibernate_go_dormant(now);
  810. } else {
  811. log_notice(LD_ACCT,
  812. "Accounting period ended. This period, we will hibernate"
  813. " until %s UTC",buf);
  814. }
  815. }
  816. }
  817. /** Consider our environment and decide if it's time
  818. * to start/stop hibernating.
  819. */
  820. void
  821. consider_hibernation(time_t now)
  822. {
  823. int accounting_enabled = get_options()->AccountingMax != 0;
  824. char buf[ISO_TIME_LEN+1];
  825. /* If we're in 'exiting' mode, then we just shut down after the interval
  826. * elapses. */
  827. if (hibernate_state == HIBERNATE_STATE_EXITING) {
  828. tor_assert(shutdown_time);
  829. if (shutdown_time <= now) {
  830. log_notice(LD_GENERAL, "Clean shutdown finished. Exiting.");
  831. tor_cleanup();
  832. exit(0);
  833. }
  834. return; /* if exiting soon, don't worry about bandwidth limits */
  835. }
  836. if (hibernate_state == HIBERNATE_STATE_DORMANT) {
  837. /* We've been hibernating because of bandwidth accounting. */
  838. tor_assert(hibernate_end_time);
  839. if (hibernate_end_time > now && accounting_enabled) {
  840. /* If we're hibernating, don't wake up until it's time, regardless of
  841. * whether we're in a new interval. */
  842. return ;
  843. } else {
  844. hibernate_end_time_elapsed(now);
  845. }
  846. }
  847. /* Else, we aren't hibernating. See if it's time to start hibernating, or to
  848. * go dormant. */
  849. if (hibernate_state == HIBERNATE_STATE_LIVE ||
  850. hibernate_state == HIBERNATE_STATE_INITIAL) {
  851. if (hibernate_soft_limit_reached()) {
  852. log_notice(LD_ACCT,
  853. "Bandwidth soft limit reached; commencing hibernation. "
  854. "No new connections will be accepted");
  855. hibernate_begin(HIBERNATE_STATE_LOWBANDWIDTH, now);
  856. } else if (accounting_enabled && now < interval_wakeup_time) {
  857. format_local_iso_time(buf,interval_wakeup_time);
  858. log_notice(LD_ACCT,
  859. "Commencing hibernation. We will wake up at %s local time.",
  860. buf);
  861. hibernate_go_dormant(now);
  862. } else if (hibernate_state == HIBERNATE_STATE_INITIAL) {
  863. hibernate_end(HIBERNATE_STATE_LIVE);
  864. }
  865. }
  866. if (hibernate_state == HIBERNATE_STATE_LOWBANDWIDTH) {
  867. if (!accounting_enabled) {
  868. hibernate_end_time_elapsed(now);
  869. } else if (hibernate_hard_limit_reached()) {
  870. hibernate_go_dormant(now);
  871. } else if (hibernate_end_time <= now) {
  872. /* The hibernation period ended while we were still in lowbandwidth.*/
  873. hibernate_end_time_elapsed(now);
  874. }
  875. }
  876. }
  877. /** Helper function: called when we get a GETINFO request for an
  878. * accounting-related key on the control connection <b>conn</b>. If we can
  879. * answer the request for <b>question</b>, then set *<b>answer</b> to a newly
  880. * allocated string holding the result. Otherwise, set *<b>answer</b> to
  881. * NULL. */
  882. int
  883. getinfo_helper_accounting(control_connection_t *conn,
  884. const char *question, char **answer,
  885. const char **errmsg)
  886. {
  887. (void) conn;
  888. (void) errmsg;
  889. if (!strcmp(question, "accounting/enabled")) {
  890. *answer = tor_strdup(accounting_is_enabled(get_options()) ? "1" : "0");
  891. } else if (!strcmp(question, "accounting/hibernating")) {
  892. if (hibernate_state == HIBERNATE_STATE_DORMANT)
  893. *answer = tor_strdup("hard");
  894. else if (hibernate_state == HIBERNATE_STATE_LOWBANDWIDTH)
  895. *answer = tor_strdup("soft");
  896. else
  897. *answer = tor_strdup("awake");
  898. } else if (!strcmp(question, "accounting/bytes")) {
  899. tor_asprintf(answer, U64_FORMAT" "U64_FORMAT,
  900. U64_PRINTF_ARG(n_bytes_read_in_interval),
  901. U64_PRINTF_ARG(n_bytes_written_in_interval));
  902. } else if (!strcmp(question, "accounting/bytes-left")) {
  903. uint64_t limit = get_options()->AccountingMax;
  904. uint64_t read_left = 0, write_left = 0;
  905. if (n_bytes_read_in_interval < limit)
  906. read_left = limit - n_bytes_read_in_interval;
  907. if (n_bytes_written_in_interval < limit)
  908. write_left = limit - n_bytes_written_in_interval;
  909. tor_asprintf(answer, U64_FORMAT" "U64_FORMAT,
  910. U64_PRINTF_ARG(read_left), U64_PRINTF_ARG(write_left));
  911. } else if (!strcmp(question, "accounting/interval-start")) {
  912. *answer = tor_malloc(ISO_TIME_LEN+1);
  913. format_iso_time(*answer, interval_start_time);
  914. } else if (!strcmp(question, "accounting/interval-wake")) {
  915. *answer = tor_malloc(ISO_TIME_LEN+1);
  916. format_iso_time(*answer, interval_wakeup_time);
  917. } else if (!strcmp(question, "accounting/interval-end")) {
  918. *answer = tor_malloc(ISO_TIME_LEN+1);
  919. format_iso_time(*answer, interval_end_time);
  920. } else {
  921. *answer = NULL;
  922. }
  923. return 0;
  924. }
  925. #ifdef TOR_UNIT_TESTS
  926. /**
  927. * Manually change the hibernation state. Private; used only by the unit
  928. * tests.
  929. */
  930. void
  931. hibernate_set_state_for_testing_(hibernate_state_t newstate)
  932. {
  933. hibernate_state = newstate;
  934. }
  935. #endif